━ Actual close┄ 3 components┄ 6 components┄ 9 components
Shows available closes from the year ending at the last observation, followed by 90 Fourier projection steps. T+ labels count trading observations; future dates are not assumed.
Withheld actual vs prior forecast
Each forecast is frozen before the displayed actual period begins. The observed prices in that period are excluded from model fitting.
30 / 60 / 90-session validation matrix
MAE against last-close baseline
Frequency decomposition
FFT.JS ENGINE
How to interpret the models
Each model uses up to 128 observations, endpoint linear detrending and the 3, 6 or 9 strongest eligible Fourier components, anchored to the last observed close. The charts compare 30, 60 and 90 trading-observation horizons. Dashed reconstructions of observed history are in-sample fits, not historical forecasts. Rolling-origin validation uses disjoint future evaluation windows after a 128-observation warm-up, while training histories overlap.
MAE is reported in BDT for equities and index points for DSEX. The baseline holds the last known close constant. Future outcomes are unknown. No calibrated confidence probabilities are asserted; longer horizons can be especially unstable. Excluded non-trading candles compress the equity observation axis. DSEX retains its published daily index closes and is not filtered by the equity exclusions.
Strategy evaluation
LONG / CASH
EMA 20 / 50 crossover
Signals at close → execution at next session open. 60-session warm-up. Whole shares, no leverage, 1% of preceding-20 mean volume capacity, with marked equity after estimated liquidation costs. Blocked entries retry while the bullish crossover state persists; exits may fill in portions.
Single-symbol exploratory test. No corporate-action adjustment, settlement constraints, lot size, price-limit fill model or market impact. No optimization or claim of an investable edge. Buy-and-hold comparison uses the same initial volume cap, integer shares and modeled costs; it keeps uninvested cash.
DSE history connector
GITHUB / BDSHARE
Public refresh is disabled to protect the verified cache. Scheduled refreshes must validate official records before replacing the dataset. CSV imports are available for private exploration.
MIT code license; underlying DSE data terms still apply. For local research. Not a licensed redistribution feed.
Import your market history
CSV
One instrument per file, 80–10,000 daily rows. Required columns:
date,open,high,low,close,volume
ISO dates. Plain numeric fields. No quoted fields or thousands separators. Duplicate dates and invalid candles are rejected. CSV provenance is user supplied, not independently verified.
Inspect selected raw data
Watchlists are saved in this browser. CSV imports remain in this session; reload restores verified server data.
Data provenance
Open-source components
rochi88/bdshare — actual DSE history connector, MIT. indutny/fft.js — actual FFT engine, MIT. Notices and original licenses are included in vendor/. SMC and pattern rules are original implementations; TradingView code and proprietary patterns are not scraped.
Research disclaimer
This tool provides general, experimental research, not personalized investment advice or an instruction to trade. BUY WATCH requires further confirmation; SELL REVIEW concerns existing long positions. Prices are historical and unadjusted. Corporate actions, liquidity, circuit limits, fees and execution delays can change outcomes. Backtests and Fourier forecasts do not guarantee future returns. No probability of profit is asserted. Obtain independent verification and licensed professional advice before relying on a signal. This application has not been independently certified for institutional use.